Construction and validation of a clinical prediction model for early renal injury in patients with immunoglobulin A vasculitis

Objective To construct a clinical prediction model of early renal injury in patients with Henoch-SchÖnlein purpura and verify its effectiveness.

Methods There are 165 patients with immunoglobulin A(IgA)vasculitis were selected as research objects. Patients were divided into kidney injury group (65 cases) and no kidney injury group (100 cases) according to whether the disease involved in kidney or not. Clinical data (general data including age, sex, height, body weight, heart rate, living environment, and laboratory indicators including neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio) were collected at admission. Single-factor Logistic regression analysis was used to screen the high-risk factors, and the clinical prediction model was established with the method of step by step. The model was visualized by running the rms package with R4.2.2 software. The prediction performance of the model was evaluated from three aspects: model differentiation degree, calibration degree, and extensibility (accuracy and stability). The model was internally verified by ten-fold cross-validation and Bootstrap method, and externally verified by time period validation.

Results There were statistically significant differences in living environment, heart rate, platelet-to-lymphocyte ratio, platelet count, lactate dehydrogenase, interleukin-6, urinary β₂-microglobulin and urinary microalbumin between two groups (P < 0.05). Single-factor Logistic regression analysis showed that living environment, heart rate, platelet-to-lymphocyte ratio, interleukin-6, urinary β₂-microglobulin and urinary microalbumin were the influencing factors of early kidney injury in IgAV patients (P < 0.05). A clinical prediction model for early kidney injury in IgAV patients was established based on the influencing factors, and random split 10-fold cross-validation and Bootstrap repeated sampling 1 000 times showed that the model had good accuracy and stability. The area under the curve (AUC) of receiver operating characteristic curve of the model was 0.87. Decision curve analysis showed that when the probability threshold of kidney injury in IgA vasculitis patients predicted by this model was 0.10 to 1.00, the net benefit rate of patients was greater than 0. Calibration curve analysis, Hosmer-Lemeshow goodness of fit test, and external validation all showed that the model had good predictive performance.

Conclusion The model can distinguish patients with kidney injury of IgAV in a high degree of differentiation and has a certain guiding value for clinical decision-making. The model would be perfect with multi-center data in the future.